Am I my connectome? Statistical issues in functional connectomics Brian Caffo, PhD Department of Statistics at National Cheng-Kung University, Taiwan 2015 SMART group, Department of Biostatistics Bloomberg School of Public Health, Johns Hopkins University
Acknowledgements
12:15 tomorrow
Am I my connectome? Is connectomics the key to understanding brain function? Are networkopathies the key to understanding many neurological disorders?
Struct./func. measurement (Huettel et al. 2009)
39 …………. 12 T
Voxels Time Data
Voxels Time = Mixing matrix Components Data Spatial independent Components Time Courses
Voxels Time = Components Spatial independent Components Time Courses Subject
= Yang et al. ArXiv
Time Data
Time
Homunculus: Clustering: Nebel et al. 2012
Investment in connectomics
Example studies in altered connectivity
Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds
Progress Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds Measure reproducibility in high dimensional settings Figure out how to make headway with so much noise Move away from group to individual measurements
Progress Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds Measure reproducibility in high dimensional settings Figure out how to make headway with so much noise Move away from group to individual measurements
I2C2 (Shou et al. 2013)
Graphical I2C2 (Yue et al.)
Progress Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds Measure reproducibility in high dimensional settings Figure out how to make headway with so much noise Move away from group to individual measurements
Shrinkage is a key to reproducibility
Progress Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds Measure reproducibility in high dimensional settings Figure out how to make headway with so much noise Move away from group to individual measurements
Shrinkage improvement in clustering (Mejia et al.)
Progress Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds Model based ICA Scalability Structure More general factor analytic models
Progress Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds Model based ICA Scalability Structure More general factor analytic models
Voxels Time = Mixing matrix Components Data Spatial independent Components Time Courses Mixture of normals Ying Guo (Biometrics 2011) Ani Eloyan (Biostatistics 2013) Histogram smoothing Shanshan Li (Submitted)
Progress Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds Model based ICA Scalability Structure More general factor analytic models
Voxels Time = Components Spatial independent Components Time Courses Subject
Progress Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds Model based ICA Scalability Structure More general factor analytic models
Voxels Time = Components Spatial independent Components Time Courses Subject
L = Spatial hemispheric independent Components RR RL RR L
Progress Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds Model based ICA Scalability Structure More general factor analytic models
Chen, Lindquist, Caffo, Vogelstein (in progress)
Graphs, some considerations Node definitions Population graphs Measures of graph reproducibility Promote conditional independence Graphs (as an outcome) regression Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds
Graphs, some considerations Node definitions Population graphs Measures of graph reproducibility Promote conditional independence Graphs (as an outcome) regression Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds
How do we define a population graph? (Han et al.)
Graphs, some considerations Scalability Node definitions Population graphs Measures of graph reproducibility Promote conditional independence Graphs (as an outcome) regression Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds
Graph regression (Qiu et al.)
Graphs, some considerations Node definitions Population graphs Measures of graph reproducibility Promote conditional independence Graphs (as an outcome) regression Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds
Node definition and regional averaging
Summary Better definitions Integrated imaging Testable hypotheses Longitudinal studies Statistical connectomics Reproducible measurements Intervention/causal thinking Handle on nuisances confounds
Thanks!